Speech Processing Tasks Research Articles

Neural trees: a new tool for classification

The authors propose a new classifier based on neural network techniques. The ‘network’ consists of a set of perceptrons functionally organized in a binary tree (‘neural tree’). The learning algorithm is inspired from a growth algorithm, the tiling algorithm, recently introduced for feedforward neural networks. As in the former case, this is a constructive algorithm, for which convergence is guaranteed. In the neural tree one distinguishes the structural organization from the functional organization: each neuron of a neural tree receives inputs from, and only from, the input layer; its output does not feed into any other neuron, but is used to propagate down a decision tree. The main advantage of this approach is due to the local processing in restricted portions of input space, during both learning and classification. Moreover, only a small subset of neurons have to be updated during the classification stage. Finally, this approach is easily and efficiently extended to classification in a multiclass problem. Promising numerical results have been obtained on different two- and multiclass problems (parity problem, multiclass nearest-neighbour classification task, etc.) including a ‘real’ low-level speech processing task. In all studied cases results compare favourably with the traditional ‘back propagation’ approach, in particular on learning and classification times as well as on network size.

Effect of acoustic‐cue redundancy of perception of stop‐vowel syllables by language‐delayed children

Previous studies demonstrated that language‐delayed children have difficulty processing speech signals (such as stop‐vowel syllables) which incorporate brief‐duration acoustic cues [P. Tallal and M. Piercy, Neuropsychol. 12, 83–93 (1974)]. The present study was designed to determine whether the presence of multiple cues to the place feature in stop‐vowel syllables would make a speech processing task more or less difficult for language‐delayed children. Five‐ through eight‐year‐old language‐delayed and normal children were trained to respond nonverbally to three pairs of syllables. The three pairs were phonologically the same /ba/ and /da/), but the pairs differed in the number of acoustic cues present. Pairs 1 and 2 were synthesized with two and three formants, respectively. Pair 3 was produced naturally (with multiple cues). The normal children's performance did not change significantly from one pair to another. All syllables were 250 ms in duration with 40‐ms formant transitions. However, the language‐delayed children's performance was significantly below normal for the two‐formant ba/da pair (p < 0.001), for the three‐formant ba/da pair score differences were less highly significant (p < 0.01), and for the real‐speech pair there was no significant difference between groups. These results suggest that acoustic redundancy aids language‐delayed children in responding to rapidly changing portions of the speech signal. [Work supported by NINCDS.]